基于动量BP神经网络的英文字符识别

英文字符识别已经广泛地应用于很多重要领域.已有的英文字符识别算法很多，一种典型的算法是BP神经网络算法.但是，BP神经网络算法有时不收敛，或陷入震荡.这就导致识别率下降.为此，本文研究了一种改进的称为动量BP神经网络算法用于英文字符识别.这种算法在BP神经网络算法的网络参数控制中添加一个动量系数和一个动量项.这样可以避免迭代的震荡，加快收敛速度.提高识别率.利用动量BP神经网络算法，对52个英文大小写字符进行了识别试验.实验结果表明，这种算法能获得满意的识别率.

English character recognition based on momentum BP neural network

English character recognition has been applied widely in many important areas. There are many existing methods for English character recognition. A typical one of the methods is BP neural network algorithm. However, BP neural network algorithm sometimes does not converge, or falls into oscillation. This leads to reduce the recognition rate. For this reason, in this paper, an improved algorithm, so called momentum BP neural network algorithm, is studied for English character recognition. In the algorithm, a momentum factor and a momentum term are added to the parameter control of BP neural network algorithm. This can avoid oscillation of iteration, speed up convergence, and increase recognition rate. Using the momentum BP neural network algorithm, the experiments of recognizing 52 lower and upper case English characters have been carried out. The experimental results show that using this algorithm can obtain satisfactory recognition rate.